Date of Award

Spring 2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Rushmeier, Holly

Abstract

Creating and editing high-quality materials for photorealistic rendering can be a difficult task due to the diversity and complexity of material appearance. Material design is the process by which artists specify the reflectance properties of a surface, such as its diffuse color and specular roughness. Even with the support of commercial software packages, material design can be a time-consuming trial-and-error task due to the counter-intuitive nature of the complex reflectance models. Moreover, many material design tasks require the physical realization of virtually designed materials as the final step, which makes the process even more challenging due to rendering artifacts and the limitations of fabrication. In this dissertation, we propose a series of studies and novel techniques to improve the intuitiveness and accuracy of material design and editing. Our goal is to understand how humans visually perceive materials, simplify user interaction in the design process and, and improve the accuracy of the physical fabrication of designs. Our first work focuses on understanding the perceptual dimensions for measured material data. We build a perceptual space based on a low-dimensional reflectance manifold that is computed from crowd-sourced data using a multi-dimensional scaling model. Our analysis shows the proposed perceptual space is consistent with the physical interpretation of the measured data. We also put forward a new material editing interface that takes advantage of the proposed perceptual space. We visualize each dimension of the manifold to help users understand how it changes the material appearance. Our second work investigates the relationship between translucency and glossiness in material perception. We conduct two human subject studies to test if subsurface scattering impacts gloss perception and examine how the shape of an object influences this perception. Based on our results, we discuss why it is necessary to include transparent and translucent media for future research in gloss perception and material design. Our third work addresses user interaction in the material design system. We present a novel Augmented Reality (AR) material design prototype, which allows users to visualize their designs against a real environment and lighting. We believe introducing AR technology can make the design process more intuitive and improve the authenticity of the results for both novice and experienced users. To test this assumption, we conduct a user study to compare our prototype with the traditional material design system with gray-scale background and synthetic lighting. The results demonstrate that with the help of AR techniques, users perform better in terms of objectively measured accuracy and time and they are subjectively more satisfied with their results. Finally, our last work turns to the challenge presented by the physical realization of designed materials. We propose a learning-based solution to map the virtually designed appearance to a meso-scale geometry that can be easily fabricated. Essentially, this is a fitting problem, but compared with previous solutions, our method can provide the fabrication recipe with higher reconstruction accuracy for a large fitting gamut. We demonstrate the efficacy of our solution by comparing the reconstructions with existing solutions and comparing fabrication results with the original design. We also provide an application of bi-scale material editing using the proposed method.

COinS